Key Features
- Multi-model comparison: Test prompts across different models simultaneously.
- 65+ templates: Pre-built prompts for code generation, summarization, Q&A, and more.
- AI prompt generation: Expand basic prompts into detailed instructions automatically.
- Variables: Test prompts with multiple values using
{{variable}}syntax. - Version control: Track iterations, restore versions, collaborate with teams.
- Prompt API: Share prompts externally via API endpoints.
Create Prompts
Navigation: Agent Platform > Prompt Studio > Create Prompts.| Path | Use When |
|---|---|
| Generate a Prompt | Need AI to expand a basic instruction into a detailed prompt |
| Start from Scratch | Want full control with a blank canvas |
| Prompt Library | Starting from proven templates for common use cases |
Prompt Canvas
On the canvas, you can:- Write system prompts (model behavior) and human prompts (task instructions)
- Add variables and import CSV test data
- Generate synthetic test datasets with AI
- Define JSON schemas for structured responses
- Compare outputs across multiple models
Canvas Options
| Category | Options |
|---|---|
| Model settings | Temperature, max tokens, top-p, frequency/presence penalty |
| Output tools | Export to CSV, bookmark models |
| Prompt tools | Generate prompt, prompt library, JSON schema |
Collaboration
| Feature | Description |
|---|---|
| Sharing | Share prompts with version history preserved |
| Versioning | Commit, restore, and mark default versions |
| Roles | Owner, Editor, Viewer permissions |
| API access | Fetch prompts via versioned endpoints |
| Templates | Save prompts as reusable templates |
Best Practices
- Use version control for significant changes
- Bookmark strong-performing models
- Start with templates for common use cases
- Use variables for consistent multi-scenario testing
- Export results as CSV for analysis
- Save successful prompts as templates
Related Topics
- Models — Manage fine-tuned, open-source, and external models
- Evaluation Studio — Benchmark model performance
- Guardrails — Ensure safe AI responses